Linus Torvalds admits he has a ‘love-hate relationship with AI’

Observe ZDNET: Add us as a most popular supply on Google.

ZDNET’s key takeaways

• Torvalds likes AI, however AI typically would not like Torvalds. 
• Linux’s founder thinks there’ll at all times be work for programmers.
• AI continues to be a blended blessing on the subject of discovering and fixing safety bugs.

Talking on the Linux Basis’s Open Supply Summit North America, Linux creator Linus Torvalds stated fashionable AI instruments are reshaping how builders work on the kernel, driving up contribution quantity and exposing new social and safety stresses within the open‑supply world. However he insisted “AI is a good software, however it’s a software” reasonably than a wholesale substitute for programmers. 

Now, if solely the businesses shedding tech staff left and proper would hear. 

Additionally: Microsoft surprises with its first server Linux distribution: Azure Linux 4.0

Torvalds spoke with Verizon’s Open Supply Program Workplace Head Dirk Hohndel, who can be a Linux kernel maintainer and a pal of Torvalds’. Torvalds added that whereas the Linux kernel’s lengthy‑standing launch course of has been secure “for just about precisely 20 years” because the transfer to Git, that development broke about six months in the past as AI coding instruments took off. 

“Within the final six months, we have seen much more commits,” Torvalds famous, estimating that “the final two releases, it has been about 20% extra commits than we had within the earlier releases over a few years.”

Initially, Torvalds misinterpret the spike as pleasure round a serious model change: “At first I believed, ‘hey, individuals are excited in regards to the 7.0 launch as a result of I modified the key quantity each occasionally…’ and it seems I used to be improper. The actual change that occurred within the final six months was that the AI instruments really received adequate for lots of people… we’re seeing a particular uptick in simply growth on just about all fronts.” 

Torvalds acknowledged that the brand new instruments decrease the barrier of entry for contributors, echoing Hohndel’s commentary that “the tooling really lowers this preliminary barrier… [and] does a giant chunk of the work.” However he emphasised that the true influence is social reasonably than purely technical: “The massive ache factors in Linux, historically, and I think in most initiatives, haven’t been a lot the code itself, however… if you find yourself compelled to vary how you’re employed.”

Additionally: Ubuntu Core 26 presents an immutable Linux you possibly can belief via 2041

One of many greatest flashpoints has been the Linux kernel safety mailing record, which Torvalds stated was not too long ago “overrun by duplicate studies” generated with AI. 

“Folks suppose that after they discover a bug with AI, the primary response typically appears to be, let’s ship it to the safety record, as a result of this will have safety implications,” he stated. The end result, on a intentionally small, confidential record, was that “we have been flooded by individuals sending bugs, after which you could have this record with only a few individuals on it… and we spent all our time simply forwarding these studies to… the opposite builders who knew that space higher.”

AI and Safety

To manage, Torvalds introduced new AI safety disclosure pointers with a blunt rule: “In case you discover a safety bug with AI, it’s best to principally think about it to be public, simply because in case you discovered it with AI, 100 different individuals additionally discovered it with AI.” 

On the identical time, he urged researchers to not publish working exploits: “Relating to issues that actually are safety points, you could not need to make the exploit public… Do not be that man who then crows about it publicly and says, ‘Look, I might deliver down this large firm.'”

Torvalds linked the disclosure debate to broader shifts within the safety ecosystem. Prior to now, he stated, the kernel neighborhood would quietly notify distributions a couple of bug and ask them to improve with out detailing the vulnerability, and “more often than not, no person would determine what occurred.” Now, with AI‑accelerated evaluation, he recalled that “final week, we fastened the bug; inside three hours, there was a weblog submit in regards to the implications of that bug repair, as a result of safety individuals love getting consideration.”

Additionally: The 4th Linux kernel flaw this month can result in stolen SSH host keys

He went out of his option to argue that closing the supply is just not a solution: “I do not suppose, for instance, that the answer is to not do open supply, as a result of in case you suppose that AI cannot reverse engineer closed supply, you are in for a shock.” In reality, he warned, “closed supply is even worse on this respect, as a result of the AI cannot enable you repair the issues, however the AI certain may also help discover these issues within the first place.”

Torvalds is true. Whereas Home windows vulnerabilities, aside from the actually horrid ones, now not obtain a lot consideration, AI can be discovering loads of safety holes in Home windows as effectively. As Dustin Childs, head of menace consciousness at Pattern Micro’s Zero Day Initiative, noticed not too long ago, “Microsoft’s whole rely got here to 1,139 CVEs patched in 2025,” which was the second-highest, behind 2020. Childs expects, “as AI bugs turn out to be extra prevalent, this quantity is more likely to go larger in 2026.” 

In the meantime, again at Open Supply Summit, Hohndel criticized distributors who hype vulnerabilities with out responsibly coordinating fixes. He cited 4 latest native privilege escalation bugs within the kernel, “two of which have been disclosed precisely” with branded names, domains, and logos earlier than maintainers have been contacted. “My response is at all times, here’s a firm I by no means need to work with, as a result of in case you do this to the Linux kernel, you do that to anybody.”

Love, hate, and AI

As annoying as that is, Torvalds admitted to having a love‑hate relationship with AI. “I really actually prefer it from a technical angle. I really like the instruments. I discover it very helpful and attention-grabbing, however it’s undoubtedly inflicting ache factors,” he stated.

Additionally: 10 trillion downloads are crushing open-source repositories – this is what they’re doing about it

On the constructive aspect, he framed AI‑found bugs as “short-term ache” with lengthy‑time period advantages: “When AI finds a bug in any supply code… long run is you discovered a bug, we fastened it, that the tip result’s higher for it.” In any case, he continued, “I believe discovering bugs is nice, as a result of the true downside is all of the bugs you did not discover.”

However he warned of “social choke factors and social ache factors” as AI pours visitors into already overstretched communities, particularly within the “10s of 1000s of random initiatives that folks keep that aren’t the Linux kernel.” For small groups or solo maintainers, he stated, flood‑fashion AI bug studies may cause actual burnout, particularly when “it is a bug report, and once you ask for extra data, the individual has completed a drive-by and would not even reply your questions anymore.”

Torvalds added that upkeep is more and more about individuals reasonably than code. “For me, as a top-level maintainer, I do not do plenty of coding. My job is working with individuals, and I don’t use AI to work with individuals. Thanks. And I ought to counsel you do not do this both.” Torvalds has come a good distance from the times when he was identified for treating poor coders with contempt. 

The way forward for AI and programming work

Stepping away from Linux, when requested what recommendation he would give to somebody originally of their profession amid doom‑and‑gloom forecasts that “all code might be written by AI,” Torvalds pushed again exhausting on advertising and marketing claims. 

“My opinion has at all times been that AI is a good software, however it’s a software, and once I see individuals saying, ‘hey, 99% of our code is written by AI,’ I actually get indignant.” 

He contrasted these claims with the fact that “100% of their code is written by compilers,” and traced his personal path from hand‑entered machine code to assemblers, then compilers, and now AI helpers. “I grew up writing machine code, and once I say machine code, I do not imply meeting language, I imply the numbers,” he stated, recalling that “it took me some time to know that writing down the numbers and calculating offsets for branches is sort of silly, and other people had give you this software known as an assembler, after which in a while I found out compilers are good too. Today, I am determining AI instruments are good too.”

So, Torvalds argued, “I am personally 100% satisfied that AI is altering programming, however it’s not altering the basics.” Simply as compilers elevated productiveness “by an element of 1000,” he estimates that “AI will improve your productiveness by an element of 10,” however insists “AI is nice, however AI is just not altering programming.” 

As a substitute, he contended, “lots of people will use AI to generate the code that the compilers use to generate the code that the assemblers then use to generate the machine code. That is revolutionary in the identical sense that we have seen revolutions earlier than.”

Crucially, Torvalds stated, would‑be builders nonetheless want to know what their instruments produce. “You do need to perceive the way it all works in the long run,” he stated. “Even once I use AI for my pet toy initiatives, I’ll use AI to generate code, I’ll take a look at that code, I’ll really nonetheless take a look at the meeting language… as a result of it is what I grew up with.” For any critical, lengthy‑lived system, he warned, “it is advisable perceive not simply your prompts, however it is advisable perceive the tip end result too, as a result of that is the one approach you possibly can keep it long run.”

Additionally: 51% of pros say AI workslop lowers their productiveness – cease it in 2 steps

All through the session, Torvalds returned to a constant theme: open supply and now AI instruments are highly effective methods to handle software program complexity, however they don’t exchange the necessity for human judgment, neighborhood norms, and a deep understanding of the methods being constructed. 

“Software program may be very sophisticated,” he stated, and “the one actually good option to handle the complexity of a fancy infrastructure is open supply,” with AI now layered in as only one extra software within the programmer’s toolbox.